On 11/14/2005 4:39 PM, Andrew Sullivan wrote:

> On Mon, Nov 14, 2005 at 01:17:11PM -0500, Christopher Browne wrote:
>> Responding to error messages is tougher than "expecting things to
>> work," that's all...
> 
> Not if your error messages are predictable, I think.

This is only valid to some degree.

Things in a slony regression test are a little more difficult than in 
the Postgres ones. Postgres regression tests contain a series of 
commands that are executed serialized. So the "entire" output is not 
only predictable about the single message, but also their order.

This isn't so simple in a replication system where you want to test that 
on a failover a non-forwarding direct subscriber gets lost if he happens 
to be ahead of all other subscribers. I think we don't even guard in any 
way against this case yet, but that's not the question here right now. 
My point is that the log messages are relatively useless for the purpose 
of checking the outcome of the regression test. The intermix of config 
events and sync's will unlikely be the same between two runs. So what 
you will have to do is to grep for certain messages only and see that 
those are the right ones in the right order.


Jan

-- 
#======================================================================#
# It's easier to get forgiveness for being wrong than for being right. #
# Let's break this rule - forgive me.                                  #
#================================================== JanWieck at Yahoo.com #